Multi-Region / Multi-Account Failover
Same Bedrock model deployed in two regions with automatic failover, or two AWS accounts with different IAM postures for blast-radius isolation.
💡
Tip: Region failover is the first real reliability conversation after a gateway has been live for a quarter. This use case packages it as something you can run, not just diagram.
What this use case covers
- The same Bedrock model deployed in two AWS regions, fronted by a single LiteLLM alias.
- Automatic failover when one region degrades — circuit-breaker thresholds, recovery probes, and what counts as "degraded."
- Two AWS accounts with different IAM postures, for blast-radius isolation between environments or tenants.
- The trade-off between active-active, active-passive, and regional pinning — and when each one is the right call.